Member
Joined 2003
I'm not exactly sure of the comparison that's being attempted here. I'd argue that your results show rather good correlation when comparison is actually equal, such as sine sweep to sine sweep. PN is a very different stimulus, its nonsensical to try and compare it's result to a sine seep. You also show that too short of a window is problematic, which should be fairly obvious. What is great about FSAF is that it captures time domain issues that are more often than not completely missed in frequency domain tests such as basic HD or even a multitione test, and you can hear it for yourself as well as analyze it in the time domain. Hear that "snap, crackle, pop" during playback? It's not a constant signal mode that's captured easily in a constant tone test, and it's broad spectrum so it affects the overall spectral content of the residual significantly. The reality is that when FSAF shows you some thing more than what you see in single tone or multitone tests, that more is going on in the measured audio that is not being captured in the other tests, whether it's background noise, room reverb, resonance, or some time domain specific occurrence like barkenhousen noise. Downside of FSAF is that it's very susceptible to background noise of any sort, so you have to treat the measurement conditions for this test like a studio recording environment.
From the residual audio you've provided, I would suggest that room reverb is likely the dominating factor. For detailed troubleshooting of the recording, measure with "save residual, mic input, and excitation" selected, so all 3 files can be compared, original audio, what the mic recorded, and residual audio.
Viewing the un-smoothed residual spectrum is nonsensical, you can just view the fundamental on the distortion tab, with SPL or dBFS units selected to show it properly. But, since it appears you've found the "Load FSAF residual" button, next step is to make your way to the spectrogram tab for detailed analysis. IMO the overall FSAF residual chart that REW shows in the distortion tab is a bit over-simplistic view of what's going on. A lot happens in 10-15s of audio that is not easily summed up in a single spectrum chart.
FWIW REW is also very powerful for multitone tests. Use the generator and RTA for this. Set up the multitone stimulus in the generator, and record with the RTA with "show distortion" selected. When you save the measurement, view in the distortion tab to see the residual with the fundamental tones removed.
Obviously.
What's been learned is that FSAF in REW recorded increased distortion across the board, with all types of signal and settings, in the higher SPL measurements - which is what was expected, and verifies at least the most basic functionality.
However, FSAF with PN accurately reproducing part of the DUT's pre-established distortion characteristic (high frequency region - observe closely) in the lower SPL run, while showing basically a flat line in the higher SPL run, is still extremely suspect. Room influence doesn't change with SPL, and there are no other environmental factors, or of the speaker, which would explain such a behavior.
Burden of proof is still on the party making a claim, so I never have to prove something doesn't work, when it hasn't been demonstrated to work in the first place. Not that I'd want to though, actually - as I've said, I'd love for FSAF to work reliably and easily.
As for IR length, I cite the REW documentation: "Choosing a length that is much longer than the time it takes the IR to decay into the noise floor may result in a processing error due to numerical ill-conditioning." At the minimum value of IR length, 100ms, that's already right around where that happens. So I do not at all agree that that's "obviously" too short a window.
I'll agree that the remainder of the room influence, as low as it'll be in a semi-NF measurement, might still be negatively influencing the results. I decided to, whenever I do free field measurements again, throw in another FSAF comparison - @tktran303 can confirm, I sent him a PM to that regard even before my first post ITT.
FWIW, I've seen you drooling over the possibility to analyze the residual as a spectrogram over at AVNirvana. I am, and have been, well aware of that option.
I've performed multitone measurements in REW before, out of curiosity, to compare the results to ARTA. Works fine - correlation is high. REW also offers even greater functionality / features. I am well aware of it, and I still chose ARTA for the comparison measurement, because, like I said, routine.
So, I'm afraid you have taught me nothing - except perhaps that complaints about the lack of empirical testing can quickly turn into the opposite, when they both far exceed your own testing, as well as don't confirm your treasured assumptions.
Have a good one.
Member
Joined 2003
Okay, good-bye.
By the way, the two residual files you posted labelled 110dB and 125dB are identical.
Last edited:
Fanatastic,
Thank you.
I always thought there was a good reason for a 4 way speaker… or a 3 way speaker with (a properly) integrated subwoofer(s)
I believe this is good evidence of this @5th element
For indoor testing with a long RT60, I prefer log chirp/exponential sine sweep, or STEPS over FSAF.
But FSAF has it uses, and for now is in beta testing phase. So please, let us not shoot the messenger(s).
We are working with it to see how, when, or why it is a useful adjunct, and how it can help (or not) with our goals. e.g. @jackocleebrown of KEF
For me, decades ago the audiophiles already thought that a 4 way speaker the size of your car was good thing.
Maybe they were right, prior to any evidence (apart from their senses)
Or the single full range driver fans also had a point, when they played a particular style of music (ie. limited bass) or had a particular type of crossover (if it’s a bad 2 way crossover, it might be better to have no crossover) or had amps with high output impedance and capable of 8W…
Last edited:
Hi everyone,
I don’t mean to hijack the thread—just hoping to contribute something that might be of interest to those following this discussion.
If you're into FSAF and its ability to use music as a test signal for measurements, you might also want to check out this related thread I started:
👉 A New/Old Measurement Tool
It covers a lesser-known technique developed years ago by the Italian magazine Audioreview, aimed at improving the correlation between loudspeaker measurements and perceived sound quality—again, with the interesting possibility of using actual music as the test signal.
The method was developed with the theoretical support of Prof. Francesco Romani, Professor of Algorithms and Data Structures at the University of Pisa, and it was later referenced in a 2009 AES paper titled "Silence Sweep: A Novel Method for Measuring Electro-Acoustical Devices" by Angelo Farina, Department of Industrial Engineering, University of Parma.
The full set of technical articles has recently been republished online (in Italian). In my thread, I’ve provided an English summary and partial translation, along with a free Java-based measurement tool (TND – Total Noise Distortion) created by a reader of the magazine.
Hope it adds something useful to the conversation!
Cheers,
Roland
I don’t mean to hijack the thread—just hoping to contribute something that might be of interest to those following this discussion.
If you're into FSAF and its ability to use music as a test signal for measurements, you might also want to check out this related thread I started:
👉 A New/Old Measurement Tool
It covers a lesser-known technique developed years ago by the Italian magazine Audioreview, aimed at improving the correlation between loudspeaker measurements and perceived sound quality—again, with the interesting possibility of using actual music as the test signal.
The method was developed with the theoretical support of Prof. Francesco Romani, Professor of Algorithms and Data Structures at the University of Pisa, and it was later referenced in a 2009 AES paper titled "Silence Sweep: A Novel Method for Measuring Electro-Acoustical Devices" by Angelo Farina, Department of Industrial Engineering, University of Parma.
The full set of technical articles has recently been republished online (in Italian). In my thread, I’ve provided an English summary and partial translation, along with a free Java-based measurement tool (TND – Total Noise Distortion) created by a reader of the magazine.
Hope it adds something useful to the conversation!
Cheers,
Roland
That can't be emphasized enough. It's per definition "work in progress - to be tested by the public" for now.
I added 25 dB gain for both SPLs / files (just before clipping for the louder one), and cut off the silence in the beginning, to make the samples easier to listen to. Something must have gone wrong in Audacity. The two files in the .rar in post #196 are indeed identical. Attached is an updated .rar with the correct(ed) files:
Last edited:
Member
Joined 2003
FSAF music stimulus residual:
I don't know...it looks fairly "well correlated" to me. The lack of change in distortion amplitude in the top end with music signal is likely because simply of lacking acoustic energy, compared to what appears to be a white spectrum stimulus used for IMD. You really need to avoid viewing FSAF residual as relative to fundamental dBr scale, view absolute amplitude and perhaps it will be clearer. Or don't listen to me, whatever you choose, I've got nothing to gain or to prove from this, just trying to get DIYers excited about this type of evaluation and what can be learned from it. Maybe it's nothing, we learned all we needed to know a decade ago.
For noise comparison to music spectrum, I'd recommend using M-Noise instead of pink noise.
And make the anechoic chamber out of concrete blocks and seal it from the rest of the universe, noise will be as much of a problem as reflections. Averaging will help some but as each doubling of takes only gets you 3dB, averages get astronomical pretty fast.
My more SNR sensitive measurements, such as low level multitone measurements, I perform in locations far from civilisation (no machine & people noise), at night (no birds chirping), on wind free days, with a high pass set on my mic pre-amp (because ambient & electronics noise is usually LF dominant). And still, an issue that would regularly pop up is planes flying overhead - even if dozens or hundreds of km away, their noise would still ruin the measurement - I would have to pause my measurement routine for a couple of minutes until they pass.
As for "in the wilderness at night", I should add that I live in Austria, which is one of the only places on Earth in which this is possible. Worst wildlife here are foxes and boars.. and they stay away. Rabies has been eradicated. We have very low crime as well. In many to most other places, you'll run the risk of, well, getting eaten 🙂 Or stabbed. Or both 🙂
If the blocks are hollow sand is your friend. Or if you are handy you can make a dual wall wooden box with a sand core. There's other tricks too. That pyramidal foam is very narrow band influence on the interior. There are much better wideband absorbers of sound.
Mark
"Wait! Can I just see how this is going to work out before I go? Here, look at this 'before' graph..." As soon as you see the eyes move, that's when you sucker-punch! And run...